Turn spoken text into written prose.
Speech tends to be conversational ("um... so... like, the thing is..."). With AI Tidy on, that gets auto-rewritten into structured written prose — great for emails, docs, and reports.
What AI Tidy does
The recognition engine handles "voice → text." AI Tidy is the next step: rewrite raw recognized text into something more suitable for writing. Specifically:
- Removes filler words ("um", "uh", "you know", "like", "so")
- Adds missing punctuation
- Fixes homophone errors based on context
- Breaks run-on sentences into shorter clauses for clarity
- Won't invent content not in your speech (hard constraint)
If a tidy doesn't succeed (e.g. an unstable network), VoiceInput simply inserts the raw recognized text at your cursor — you never lose words to a failed tidy, and never wait on a spinner.
Toggle
Settings → Tidy → Enable AI Tidy
When on, every recording goes through AI Tidy. Skip just this once: double-tap right Option (raw recognition only).
AI source · two choices
Pick which service powers AI Tidy under Settings → AI Model → AI Source:
- VoiceInput Cloud AI Tidy (recommended): VoiceInput's cloud tidy service — text appears the moment you release the key, fast and accurate, works out of the box with zero configuration. The right choice for almost everyone.
- Your own API key: use your own OpenAI-compatible API key — unlimited, permanently free, no subscription needed. For advanced users.
Quota rules
- Free: 50 Cloud AI Tidy calls per day (resets daily)
- Pro: unlimited Cloud AI Tidy
- Your own API key: no quota at all — calls go through your own key, unrelated to our limits
Your own API key (BYOK)
BYOK = Bring Your Own Key. VoiceInput supports any OpenAI-compatible provider — as long as it offers an OpenAI-format API (almost every major LLM provider does), just enter the API key and endpoint.
How to configure
Settings → AI Model → AI Source → Your own API key
Expand "Advanced" → enter API key and model → test connectivity → save.
API keys are stored in macOS Keychain (system-level encryption) — never uploaded anywhere. Uninstalling VoiceInput clears the keys from Keychain.
Hard constraints — what AI Tidy never does
These are baked into the system prompt:
- Never invent content you didn't say (no "completion")
- Never change meaning (no "optimization" into something else)
- Never translate languages (Chinese stays Chinese)
- Never answer your questions (you say "what time is it" → it returns that text, doesn't tell you the time)
If you ever see AI Tidy do any of the above, it's a bug — please report to [email protected].